Research map / Architectures and efficiency / RNNs and normalization
APIC: Orthogonalized Neuro-Symbolic Modeling for Nonlinear Dissipative Dynamics
YanHui Zhu, Xiangfu Meng, Chen Zhao, Yinhao Li
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add gated fusion of a Navier-Stokes physics branch (ablation) | data-driven backbone · 3D CFD, 5% train data | L2: 0.0865 -> 0.0794 | Improved |
| replace linear gated fusion with stacked (concatenation + nonlinear) fusion (ablation) | APIC with NS operator · 3D CFD, 5% train data | L2: 0.0794 -> 0.0646 | Improved |
| replace Navier-Stokes operator with G-KSCH operator basis (ablation) | APIC stacked fusion · 3D CFD, 5% train data | L2: 0.0646 -> 0.0586 | Improved |
| add stop-gradient (gradient isolation) on physics prediction (ablation) | APIC · 3D CFD, 5% train data | L2: 0.060 (no-detach) vs 0.062 (detach) | No clear effect |
| add stop-gradient (gradient isolation) on physics prediction (ablation) | APIC · 3D CFD, 10% train data | L2: 0.050 vs 0.050 | No clear effect |
| add physics branch to data backbone (ablation) | data-only backbone · 3D CFD, 5% train data | L2: 0.093 -> 0.062 | Improved |
| train streams in separate stages then fit adapter, instead of joint training (ablation) | APIC · 3D CFD, 5% train data | L2: 0.062 -> 0.114 | Hurt |
| use physics branch alone without data residual backbone (ablation) | APIC · 3D CFD, 5% train data | L2: 0.062 -> 0.334 | Hurt |
| add reaction term in traffic operator (ablation) | APIC (Traffic) · TaxiBJ | RMSE: 18.097 -> 17.810 | Improved |
| add congestion potential term (ablation) | APIC (Traffic) · TaxiBJ | RMSE: 17.959 -> 17.810 | Improved |
| adapt operator basis with rotation term and decoupled coefficients (ablation) | APIC · WeatherBench | RMSE: 18.26 -> 16.88 | Improved |
Closest papers by meaning
- Universal Physics Transformers: A Framework For Efficiently Scaling Neural Operators (NeurIPS 2024 · similarity 0.86)
- Neuro-Spectral Architectures for Causal Physics-Informed Networks (NeurIPS 2025 · similarity 0.86)
- Stable Port-Hamiltonian Neural Networks (NeurIPS 2025 · similarity 0.86)
- Spectral-Refiner: Accurate Fine-Tuning of Spatiotemporal Fourier Neural Operator for Turbulent Flows (ICLR 2025 · similarity 0.85)
- OrthoReg: Orthogonal Regularization for Hybrid Symbolic-Neural Dynamical Systems (NeurIPS 2026 · similarity 0.85)
- Hierarchical Implicit Neural Emulators (NeurIPS 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
